BUAA-MSOD Dataset: Multiple Moving Space Object Detection Benchmark

BUAA-MSOD (Beihang University Astronomical Multiple Space Object Dataset) is a dedicated benchmark designed to support research on accurate multi-target detection in wide-field astronomical imaging.

Overview

The dataset is derived from real observational image sequences captured by ground-based optical telescopes. It features spaceborne targets exhibiting:

- Diverse morphologies  
- Varying motion patterns  
- Realistic noise and imaging conditions

Data Acquisition

- Observation site: Xinglong Observatory, National Astronomical Observatories, Chinese Academy of Sciences  
- Telescope mode: Track rate  
- System: Wide-field surveillance system  
- Field of view: 5° × 5°  
- Exposure times: 150 ms and 240 ms  
- Resolution: 16-bit TIFF, 6k × 6k pixels  
- Sequences: 4 observational sequences  
- Frames per sequence: 60 consecutive images  

Annotation and Preprocessing

We manually annotated four groups of sequential images and performed standardized data preprocessing. The final detection dataset reflects various space object trajectories and motion behaviors.

Dataset Split

| Subset       | Images | Labeled Targets |
|--------------|--------|------------------|
| Training     | 3,490  | 1,862            |
| Validation   | 3,840  | 366              |
| Test         | 3,840  | 355              |

Applications

The BUAA-MSOD dataset serves as a valuable resource for:

- Multi-object detection in astronomical images  
- Space object tracking and trajectory analysis  
- Benchmarking motion-aware models under wide-field settings  

---

Related Resources

- [Code Repository (GitHub)](https://github.com/yx-gg/MSAMNet)

Please cite this dataset if used in your research.